Convolutional neural network-based prediction of void evolution by hydrogen-helium synergistic effect
[Objective]The hydrogen-helium synergistic effect can further enhance irradiation damage and exacerbate the irradiation-induced degradation of the physical and mechanical properties of structural materials.However,due to the time-and resource-intensive property of both experiments and multi-scale simulations in irradiation damage,the trial-and-error approach is entirely inefficient.It is therefore important to explore new methods to accelerate the development of potentially irradiation-resistant materials.In this paper,void size and number density models based on convolutional neural networks(CNN)have been developed to systematically predict the correlation of irradiation parameters across a range of parameter variations.[Methods]Deep learning techniques such as adaptive moment estimation(Adam),dropout layers,and batch normalization,are applied to improve the generalization ability of the prediction models.The model performance is further optimized in three aspects:activation function,number of convolutional layers,and dropout layer values.Finally,the prediction of void size and number density is carried out within a continuous parameter variation range.[Results]Compared with the Sigmoid and Tanh activation functions,the error reduction of the Relu is smoother and exhibits higher computational efficiency.When the number of convolution layers is 3 and the dropout layer value is 0.2,the models achieve the best performance under three evaluation indicators:root mean square error,coefficient of determination(R2),and mean absolute error.The R2 of the size prediction model is 92.40％and 88.80％for the training set and testing set,respectively,indicating the absence of underfitting or overfitting problems.Although there is a slightly poor fitting effect on the testing set between the predicted results of the number density prediction model and the measured values,indicating an overfitting problem that needed to be solved,the overall prediction results are relatively consistent with the measured values.The predicted results show that,the void size and number density increase with increasing irradiation damage,suggesting a linear dependence.The void size is small and the number density is large under double-beams irradiation(heavy ions+He+),while the co-injection of He and H under triple-beams irradiation(heavy+He++H+)strongly promotes the void growth.Meanwhile,there is a peak in void size with increasing temperature,and the growth of the void is the result of a sharp decrease in number density.Finally,there is a critical value between the He and H atom injection rates and the void size and number density,suggesting that the evolution of the void under the synergistic effects is controlled by a combination of different He and H atom injection rates.[Conclusions]In conclusion,CNN,as an alternative method,can predict void evolutions without involving physical processes and corresponding physics-based modeling techniques.There are significant synergistic effects between helium,hydrogen,and displacement damage,contributing to void growth rather than nucleate.Therefore,the most crucial aspect in the development of anti-swelling materials is to suppress the growth of voids.In addition,the predicted results can offer inspiration and reference for the design and optimization of high-performance irradiation-resistant materials.For example,the maximum or minimum He and H atom injection rates do not induce the maximum void size,and material degradation may be mitigated by optimizing irradiation parameters.The performance prediction of such systems is difficult to accomplish with experimental methods,but it holds important implications for material design and performance assessment.
万方数据
维普
